To be honest, I don't yet feel fully confident in anything here. The latest upload builds on armel/armhf without your patch and I'm hoping debci will pass too, suggesting that Helge's patch is really the right solution.
/Simon
Drew Parsons <dpars...@emerall.com> writes:
> Thanks for following up the patch, Helge. This sounds more like
> Bug#1086769, which is closed but not yet archived.
>
> Simon Josefsson is the main maintainer for uid-wrapper, so he's in a
> better position to judge if sskip-arm-32bit-syscall.patch should be
> updated. I'm forwarding your message to Bug#1086769 and him to review
> further before #1086769 get archived.
>
> On your side Helge, can you check if Simon's recent changes in 1.3.1-2
> address the issue you've raised here?
>
> Cheers,
> Drew
>
>
> On 2024-11-05 18:15, Helge Deller wrote:
>> Hi Drew,
>> I noticed this bug report is "archived", so I don't know if
>> my info will get attached to the report.
>> Since you came up with the sskip-arm-32bit-syscall.patch,
>> wich I believe is wrong, could you please check my patch below?
>> Helge
>> -------- Forwarded Message --------
>> Subject: Re: Bug#1069425: uid-wrapper: FTBFS on armhf: 15: [ LINE ]
>> --- ./tests/test_syscall.c:53: error: Failure!
>> Date: Tue, 5 Nov 2024 18:10:53 +0100
>> From: Helge Deller <del...@gmx.de>
>> To: 1069...@bugs.debian.org
>> The reason test15 (test_uwrap_syscall) fails on some 32-bit
>> platforms is due to the time64 transition:
>>> 15: [ RUN ] test_uwrap_syscall
>>> 15: [ ERROR ] --- 0x66234648 != 0x42c0c66234648
>>> 15: [ LINE ] --- ./tests/test_syscall.c:53: error: Failure!
>>> 15: [ FAILED ] test_uwrap_syscall
>> Because of the transition, "-D_LARGEFILE_SOURCE
>> -D_FILE_OFFSET_BITS=64"
>> gets added and as such the "struct timeval tv2" gets extended to 64bit
>> which does not fit the struct layout (and size) which the syscall
>> expects.
>> See usr/share/perl5/Dpkg/Vendor/Debian.pm, line ~303:
>> # In Debian gcc enables time64 (and lfs) for the following
>> architectures
>> # by injecting pre-processor flags, though the libc ABI has not
>> changed.
>> if (any { $arch eq $_ } qw(armel armhf hppa m68k mips mipsel
>> powerpc sh4)) {
>> $flags->set_option_value('cc-abi-time64', 1);
>> The uid-wrapper is a very low-level package and as such it should
>> not
>> be built with time64 enabled (same as glibc is built without it).
>> The obvious fix for this package is the following trivial patch.
>> I tried hppa, but it should solve the issue for the other architectures
>> as well: armel armhf hppa m68k mips mipsel powerpc sh4
>> diff -up ./debian/rules.bak ./debian/rules
>> --- ./debian/rules.bak 2024-11-05 16:53:45.182697375 +0000
>> +++ ./debian/rules 2024-11-05 16:04:57.142488073 +0000
>> @@ -1,6 +1,6 @@
>> #!/usr/bin/make -f
>> -export DEB_BUILD_MAINT_OPTIONS = hardening=+all
>> +export DEB_BUILD_MAINT_OPTIONS = hardening=+all abi=-time64
>
signature.asc
Description: PGP signature
